Network Working Group                                          W. Kumari
Request for Comments: 5635                                        Google
Category: Informational                                     D. McPherson
                                                          Arbor Networks
                                                             August 2009

Remote Triggered Black Hole Filtering
with Unicast Reverse Path Forwarding (uRPF)

Abstract

   Remote Triggered Black Hole (RTBH) filtering is a popular and
   effective technique for the mitigation of denial-of-service attacks.
   This document expands upon destination-based RTBH filtering by
   outlining a method to enable filtering by source address as well.

Status of This Memo

   This memo provides information for the Internet community.  It does
   not specify an Internet standard of any kind.  Distribution of this
   memo is unlimited.

Copyright Notice

   Copyright (c) 2009 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents in effect on the date of
   publication of this document (http://trustee.ietf.org/license-info).
   Please review these documents carefully, as they describe your rights
   and restrictions with respect to this document.

   1. Introduction ....................................................2
   2. Terminology .....................................................3
   3. Destination Address RTBH Filtering ..............................3
      3.1. Overview ...................................................3
      3.2. Detail .....................................................4
   4. Source Address RTBH Filtering ...................................7
      4.1. Steps to Deploy RTBH Filtering with uRPF for Source
           Filtering ..................................................8
   5. Security Considerations .........................................9
   6. Acknowledgments .................................................9
   7. References ......................................................9
      7.1. Normative References .......................................9
      7.2. Informative References ....................................10
   Appendix A. Cisco Router Configuration Sample .....................11
   Appendix B. Juniper Configuration Sample ..........................12
   Appendix C. A Brief History of RTBH ...............................14

1. Introduction

   This document expands upon the technique outlined in "Configuring BGP
   to Block Denial-of-Service Attacks" [RFC3882] to demonstrate a method
   that allows for filtering by source address(es).

   Network operators have developed a variety of techniques for
   mitigating denial-of-service (DoS) attacks.  While different
   techniques have varying strengths and weaknesses, from an
   implementation perspective, the selection of which method to use for
   each type of attack involves evaluating the tradeoffs associated with
   each method.

   A common DoS attack directed against a customer of a service provider
   involves generating a greater volume of attack traffic destined for
   the target than will fit down the links from the service provider(s)
   to the victim (customer).  This traffic "starves out" legitimate
   traffic and often results in collateral damage or negative effects to
   other customers or the network infrastructure as well.  Rather than
   having all destinations on their network be affected by the attack,
   the customer may ask their service provider to filter traffic
   destined to the target destination IP address(es), or the service
   provider may determine that this is necessary themselves, in order to
   preserve network availability.

   One method that the service provider can use to implement this
   filtering is to deploy access control lists on the edge of their
   network.  While this technique provides a large amount of flexibility
   in the filtering, it runs into scalability issues, both in terms of
   the number of entries in the filter and the packet rate.

   Most routers are able to forward traffic at a much higher rate than
   they are able to filter, and they are able to hold many more
   forwarding table entries and routes than filter entries.  RTBH
   filtering leverages the forwarding performance of modern routers to
   filter more entries and at a higher rate than access control lists
   would otherwise allow.

   However, with destination-based RTBH filtering, the impact of the
   attack on the target is complete.  That is, destination-based RTBH
   filtering injects a discard route into the forwarding table for the
   target prefix.  All packets towards that destination, attack traffic
   AND legitimate traffic, are then dropped by the participating
   routers,thereby taking the target completely offline.  The benefit is
   that collateral damage to other systems or network availability at
   the customer location or in the ISP network is limited, but the
   negative impact to the target itself is arguably increased.

   By coupling unicast Reverse Path Forwarding (uRPF) [RFC3704]
   techniques with RTBH filtering, BGP can be used to distribute discard
   routes that are based not on destination or target addresses, but on
   source addresses of unwanted traffic.  Note that this will drop all
   traffic to/from the address, and not just the traffic to the victim.

   This document is broken up into three logical parts: the first
   outlines how to configure destination-based RTBH, the second covers
   source-based RTBH, and the third part has examples and a history of
   the technique.

2. Terminology

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED",  "MAY", and "OPTIONAL" in this
   document are to be interpreted as described in RFC 2119 [RFC2119].

3. Destination Address RTBH Filtering

3.1. Overview

   A discard route is installed on each edge router in the network with
   the destination set to the discard (or null) interface.  In order to
   use RTBH filtering for a single IP address (or prefix), a BGP route
   for the address to be filtered is announced, with the next-hop set as

   the discard route.  This causes traffic to the announced network
   prefix to be forwarded to the discard interface so that it does not
   transit the network wasting resources or triggering collateral damage
   to other resources along the path towards the target.

   While this does "complete" the attack in that the target address(es)
   are made unreachable, collateral damage is minimized.  It may also be
   possible to move the host or service on the target IP address(es) to
   another address and keep the service up, for example, by updating
   associated DNS resource records.

3.2. Detail

   Before deploying RTBH filtering, there is some preparation and
   planning that needs to occur and decisions that need to be made.
   These include:

   -  What are the discard addresses?

   -  What are the discard BGP communities?

   -  What is the largest prefix that can be black-holed?

   -  What is the smallest advertisement that your provider will accept?

   Steps to configure destination-based RTBH filtering:

   Step 1. Select Your Discard Address Schema

   An address is chosen to become the "discard address".  This is often
   chosen from 192.0.2.0/24 (TEST-NET [RFC3330]), or from RFC 1918
   [RFC1918] space.  Multiple addresses allow an operator to configure
   multiple static routes, one for each incident:

      192.0.2.1 = Incident #1
      192.0.2.2 = Incident #2
      192.0.2.3 = Incident #3
      192.0.2.4 = Incident #4
      192.0.2.5 = Incident #5

   Customer #1, who has a DDoS (Distributed DoS) attack can be pointed
   to discard route 192.0.2.1.  Customer #2 can be pointed to discard
   route 192.0.2.2.  If capable, the router can then count the drops for
   each, providing some level of telemetry on the volume of drops as
   well as status of an ongoing attack.  A consistent address schema
   facilitates operations.

   Step 2. Configure the Discard Route(s) on Each Router

   A route for the "discard address" is installed on the routers that
   form the edge/perimeter of the network in all routers in the network
   or some subset (e.g., peering, but not customer, etc.).  The
   destination of the route is set to the "discard" or "null" interface.
   This route is called the "discard route".  Implementation experience
   demonstrates the value of configuring each ingress router with a
   capability for dropping traffic via RTBH filtering.

   Step 3. Configure the RTBH BGP Policy on Each Router

   A BGP policy is configured on all routers that have the discard route
   so that routes announced with a chosen community will have their
   next-hop set to the discard address.  The BGP policy should be made
   restrictive so that only BGP routes covering a defined number of
   hosts addresses will be accepted.  That is, typically, only specific
   /32s are necessary.  Shorter prefix blocks may also be required or
   desirable, for example, if larger numbers of attack targets are
   located within a single prefix, or the employment of this mechanism
   is to drop traffic bound for specific networks.  When filtering based
   on shorter prefixes, extreme caution should be used as to avoid
   collateral damage to other hosts that reside within those address
   blocks.  Full implementations will have multiple communities, with
   each community used for different parts of a provider's network and
   for different security incidents.

   Step 4. Configure the Safety Egress Prefix Filters

   There is always a chance that the triggering BGP update could leak
   from the network and so egress prefix filters are strongly
   recommended.  These egress prefix filter details may vary, but
   experience has demonstrated that the following works:

   -  Deny all prefixes longer than the longest prefix that you expect
      to announce.  For example, if the longest prefix that you expect
      to announce is /24, deny all prefixes of length /25 though /32.
      If your triggering BGP update is only /32s, then this egress
      prefix filter will add a safe measure in case the NO_EXPORT
      community does not work.

   -  Deny all communities used for triggering RTBH filtering.  This is
      also a "safety" measure in case the NO_EXPORT community does not
      work.

   Step 5: Configure the Trigger Router

   Configure the trigger router, workstation, or other device so that
   adding and removing the triggers can be done easily and quickly.  The
   BGP update should have the NO_EXPORT community as a mandatory
   attribute.  An egress prefix filter or policy that prevents RTBH
   filtering prefixes in the /8 to /24 range is also recommended as a
   safety tool.  The trigger router can be set up as an iBGP (Internal
   BGP) route reflector client that does not receive any prefixes from
   its BGP peer.  This allows a low-cost router/workstation to be used
   as the trigger router.

   Using the RTBH filtering:

   1: When RTBH filtering is desired for a specific address, that
      address is announced from a trigger router (or route server),
      tagged with the chosen "RTBH" community and with the NO_EXPORT
      community, and passed via iBGP.  The receiving routers check the
      BGP policy, set the next-hop to be the discard route, resulting in
      a Forwarding Information Base (FIB) entry pointing to a discard
      address.

   2: Traffic entering the network will now be forwarded to the discard
      interface on all edge routers and will therefore be dropped at the
      edge of the network, saving resources.

      2.1: Multiple Discard Addresses for Incident Granularity.  This
           technique can be expanded by having multiple discard
           addresses, routes, and communities to allow for monitoring of
           the discarded traffic volume on devices that support multiple
           discard interfaces.  As mentioned earlier, each router can
           have a discard address schema to allow the operator to
           distinguish multiple incidents from each other -- making it
           easier to monitor the life-cycle of the incidents.

      2.2: Multiple "Trigger Communities" for Network-Wide Granularity.
           The network can be sectioned into multiple communities,
           providing the operator with an ability to drop in discrete
           parts of their network.  For example, the network can be
           divided into the following communities (where XXX represents
           the operator's AS number):

               XXX:600 RTBH filtering on all routers
               XXX:601 RTBH filtering on only peering routers
               XXX:602 RTBH filtering on only customers who peer BGP
               XXX:603 RTBH filtering on data centers (to see if the
                       data center is the source of attack)

               XXX:604 RTBH filtering on all customers (to see how many
                       customers are being used by the attacker)

           Some diligent thinking is required to develop a community
           schema that provides flexibility while reflecting topological
           considerations.

      2.3: "Customer-Triggered" RTBH filtering.  The technique can also
           be expanded by relaxing the Autonomous System (AS) path rule
           to allow customers of a service provider to enable RTBH
           filtering without interacting with the service provider's
           trigger routers.  If this is configured, an operator MUST
           only accept announcements from the customer for prefixes that
           the customer is authorized to advertise, in order to prevent
           the customer from accidentally (or intentionally) black-
           holing space that they are not allowed to advertise.

           A common policy for this type of setup would first permit any
           longer prefix within an authorized prefix only if the black
           hole communities are attached, append NO_EXPORT,
           NO_ADVERTISE, or similar communities, and then also accept
           from the customer the original aggregate prefix that will be
           advertised as standard policy permits.

           Extreme caution should be used in order to avoid leaking any
           more specifics beyond the local routing domain, unless policy
           explicitly aims at doing just that.

4. Source Address RTBH Filtering

   In many instances, denial-of-service attacks sourced from botnets are
   being configured to "follow DNS".  (The attacking machines are
   instructed to attack www.example.com, and re-resolve this
   periodically.  Historically, the attacks were aimed simply at an IP
   address and so renumbering www.example.com to a new address was an
   effective mitigation.)  This makes it desirable to employ a technique
   that allows black-holing to be based on source address.

   By combining traditional RTBH filtering with unicast Reverse Path
   Forwarding (uRPF), a network operator can filter based upon the
   source address.  uRPF performs a route lookup of the source address
   of the packet and checks to see if the ingress interface of the
   packet is a valid egress interface for the packet source address
   (strict mode) or if any route to the source address of the packet
   exists (loose mode).  If the check fails, the packet is typically
   dropped.  In loose mode, some vendors also verify that the
   destination route does not point to an invalid next-hop -- this
   allows source-based RTBH filtering to be deployed in networks that

   cannot implement strict (or feasible path) mode uRPF.  Before
   enabling uRPF (in any mode), it is vital that you fully understand
   the implications of doing so:

   -  Strict mode will cause the router to drop all ingress traffic if
      the best path back to the source address of the traffic is not the
      interface from which the traffic was received.  Asymmetric routing
      will cause strict mode uRPF to drop legitimate traffic.

   -  Loose mode causes the router to check if a route for the source
      address of the traffic exists.  This may also cause legitimate
      traffic to be discarded.

   It is hoped that in the future, vendors will implement a "DoS-
   mitigation" mode in addition to the loose and strict modes -- in this
   mode, the uRPF check will only fail if the next-hop for the source of
   the packet is a discard interface.

   By enabling the uRPF feature on interfaces at predetermined points in
   their network and announcing the source address(es) of attack
   traffic, a network operator can effectively drop the identified
   attack traffic at specified devices (ideally ingress edge) of their
   network based on source address.

   While administrators may choose to drop traffic from any prefix they
   wish, it is recommended when employing source-based RTBH filtering
   inter-domain that explicit policy be defined that enables peers to
   only announce source-based RTBH routes for prefixes that they
   originate.

4.1. Steps to Deploy RTBH Filtering with uRPF for Source Filtering

   The same steps that are required to implement destination address
   RTBH filtering are taken with the additional step of enabling unicast
   Reverse Path Forwarding on predetermined interfaces.  When a source
   address (or network) needs to be blocked, that address (or network)
   is announced using BGP tagged with a community.  This will cause the
   route to be installed with a next-hop of the discard interface,
   causing the uRPF check to fail and the packets to be discarded.  The
   destination-based RTBH filtering community should not be used for
   source-based RTBH filtering, and the routes tagged with the selected
   community should be carefully filtered.

   The BGP policy will need to be relaxed to accept announcements tagged
   with this community to be accepted even though they contain addresses
   not controlled by the network announcing them.  These announcements
   must NOT be propagated outside the local AS and should carry the
   NO_EXPORT community.

   As a matter of policy, operators SHOULD NOT accept source-based RTBH
   announcements from their peers or customers, they should only be
   installed by local or attack management systems within their
   administrative domain.

5. Security Considerations

   The techniques presented here provide enough power to cause
   significant traffic forwarding loss if incorrectly deployed.  It is
   imperative that the announcements that trigger the black-holing are
   carefully checked and that the BGP policy that accepts these
   announcements is implemented in such a manner that the announcements:

   -  Are not propagated outside the AS (NO_EXPORT).

   -  Are not accepted from outside the AS (except from customers).

   -  Except where source-based filtering is deployed, that the network
      contained in the announcement falls within the address ranges
      controlled by the announcing AS (i.e., for customers that the
      address falls within their space).

6. Acknowledgments

   I would like to thank Joe Abley, Ron Bonica, Rodney Dunn, Alfred
   Hoenes, Donald Smith, Joel Jaeggli, and Steve Williams for their
   assistance, feedback and not laughing *too* much at the quality of
   the initial versions.

   I would also like to thank all of the regular contributors to the
   OPSEC Working Group and Google for 20% time :-)

   The authors would also like to thank Steven L Johnson and Barry
   Greene for getting this implemented and Chris Morrow for publicizing
   the technique in multiple talks.

7. References

7.1. Normative References

   [RFC1918]    Rekhter, Y., Moskowitz, B., Karrenberg, D., de Groot,
                G., and E. Lear, "Address Allocation for Private
                Internets", BCP 5, RFC 1918, February 1996.

   [RFC2119]    Bradner, S., "Key words for use in RFCs to Indicate
                Requirement Levels", BCP 14, RFC 2119, March 1997.

   [RFC3330]    IANA, "Special-Use IPv4 Addresses", RFC 3330, September
                2002.

   [RFC3704]    Baker, F. and P. Savola, "Ingress Filtering for
                Multihomed Networks", BCP 84, RFC 3704, March 2004.

   [RFC3882]    Turk, D., "Configuring BGP to Block Denial-of-Service
                Attacks", RFC 3882, September 2004.

7.2. Informative References

   [Greene2001] Greene Barry Raveendran and Jarvis Neil, "Unicast
                Reverse Path Forwarding (uRPF) Enhancements for the
                ISP-ISP Edge", ftp://ftp-eng.cisco.com/
                cons/isp/documents/uRPF_Enhancement.pdf, 2001.

Appendix A. Cisco Router Configuration Sample

   This section provides a partial configuration for configuring RTBH
   filtering on a Cisco router.  This is not a complete configuration
   and should be customized before being used.

Announcing router:
   ! The discard route
   ip route 192.0.2.1 255.255.255.255 Null0
   !
   ! Matches and empty AS-PATH only.
   ip as-path access-list 10 permit ^$
   !
   ! This route-map matches routes with tag 666 and sets the next-hop
   ! to be the discard route.
   route-map remote-trigger-black-hole permit 10
    match tag 666
    set ip next-hop 192.0.2.1
    set local-preference 200
    set community no-export
    ! The community used internally to tag RTBH announcements.
    set community 65505:666
    set origin igp
   !
   route-map remote-trigger-black-hole permit 20
   !
   router bgp 65505
    no synchronization
    bgp log-neighbor-changes
    redistribute static route-map remote-trigger-black-hole
    no auto-summary
   !
   ! An example IP that we are applying RTBH filtering to.
   ! All traffic destined to 10.0.0.1 will now be dropped!
   ip route 10.0.0.1 255.255.255.255 null0 tag 666
   !

Filtering router:
   !
   ! The discard route
   ip route 192.0.2.1 255.255.255.255 Null0
   !
   ! Matches and empty AS-PATH only.
   ip as-path access-list 10 permit ^$
   !
   route-map black-hole-filter permit 10
    match ip address prefix-list only-host-routes
    match as-path 10

    match community 65505:666 no-export
   !
   ! Don't accept any other announcements with the RTBH community.
   route-map black-hole-filter deny 20
    match community 65505:666
   !
   route-map black-hole-filter permit 30
   !
   ! An interface for source-based RTBH filtering with uRPF loose mode.
   interface FastEthernet 0/0
   ip verify unicast source reachable-via any

Appendix B. Juniper Configuration Sample

   This section provides a partial configuration for configuring RTBH
   filtering on a Juniper router.  This is not a complete configuration
   and should be customized before being used.

   Announcing router:

      routing-options {
         static {
             /* EXAMPLE ATTACK SOURCE */
             route 10.11.12.66/32 {
                 next-hop 192.0.2.1;
                 resolve;
                 tag 666;
             }
             /* EXAMPLE ATTACK DESTINATION */
             route 10.128.0.2/32 {
                 next-hop 192.0.2.1;
                 resolve;
                 tag 666;
             }
         }
         autonomous-system 100;
      }

      protocols {
         bgp {
             group ibgp {
                 type internal;
                 export rtbh;
                 neighbor 172.16.0.2;
             }
         }
      }

      policy-options {
         policy-statement rtbh {
             term black-hole-filter {
                 from {
                     tag 666;
                     route-filter 0.0.0.0/0 prefix-length-range /32-/32;
                 }
                 then {
                     local-preference 200;
                     origin igp;
                     community set black-hole;
                     community add no-export;
                     next-hop 192.0.2.1;
                     accept;
                 }
             }
         }
         community black-hole members 100:666;
         community no-export members no-export;
      }

   Filtering router:

      policy-statement black-hole-filter {
         from {
             protocol bgp;
             as-path LocalOnly;
             community black-hole;
             route-filter 0.0.0.0/0 prefix-length-range /32-/32;
         }
         then {
             community set no-export;
             next-hop 192.0.2.1;
         }
      }
      community black-hole members 100:666;
      community no-export members no-export;

      routing-options {
         forwarding-table {
             unicast-reverse-path feasible-paths;
         }
         static {
             route 192.0.2.1/32 discard;
         }
      }

      interfaces {
         xe-1/0/0 {
             vlan-tagging;
             mtu 9192;
             unit 201 {
                 vlan-id 201;
                 family inet {
                     rpf-check;
                     address 10.11.12.1/24;
                 }
             }
         }
      }

Appendix C. A Brief History of RTBH Filtering

   Understanding the history and motivation behind the development of a
   technique often helps with understanding how to best utilize the
   technique.  In this spirit, we present a history of unicast RPF and
   RTBH filtering.

   This section has been provided by Barry Raveendran Greene:

   Unicast RPF Loose Check (uRPF Loose Check) was created by Neil Jarvis
   and Barry Greene to be used with destination-based RTBH as a rapid
   reaction tool to DDoS attacks.  The requirements for this rapid
   reaction tool was based on post mortem conversation after the
   February 2000 attacks on several big content hosting companies.  The
   summary of the requirement became the "Exodus Requirement" which
   stated:

      We need a tool to drop packets based on source IP address that can
      be pushed out to over 60 routers within 60 seconds, be longer than
      a thousand lines, be modified on the fly, and work in all your
      platforms filtering at line rate.

   A variety of options were looked at to meet this requirement, from
   reviving Common Open Policy Service (COPS), to pushing out Access
   Control Lists (ACLs) with BGP, creating a new protocol.  In 2000, the
   quickest way to meet the "Exodus requirement" was to marry two
   functions.  First, modify unicast RPF so that the interface check was
   no longer required and to make sure that a "null" or discard route
   would drop the packet (i.e., loose check).  Second, the technique
   where BGP is used to trigger a distributed drop is dusted off and
   documented.  Combining these two techniques was deemed a fast way to
   put a distributed capability to drop packets out into the industry.

   To clarify and restate, uRPF loose check was created as one part of a
   rapid reaction tool to DDoS attacks that "drop packets based on
   source IP address that can be pushed out to over 60 routers with in
   60 seconds, be longer than a thousand lines, be modified on the fly,
   and work in all your platforms filtering at line rate".  The
   secondary benefits of using uRPF Loose Check for other functions is a
   secondary benefit -- not the primary goal for its creation.

   To facilitate the adoption to the industry, uRPF Loose Check was not
   patented.  It was publicly published and disclosed in "Unicast
   Reverse PathForwarding (uRPF) Enhancements for the ISP-ISP Edge"
   [Greene2001].

Authors' Addresses

   Warren Kumari
   Google
   1600 Amphitheatre Parkway
   Mountain View, CA 94043

   EMail: warren@kumari.net

   Danny McPherson
   Arbor Networks, Inc.

   EMail: danny@arbor.net